Why Generative AI’s impact is still years away

Given the number of people having conversations with ChatGPT and creating quirky AI-generated art for social media, it’s easy to assume that artificial intelligence has already arrived as a major force in the American tech world. But in reality, its impact on the business landscape has been limited so far—and it may still be a few years before we see widespread, serious adoption.

Generative AI Isn’t Happening Like We’ve Been Told, a report from Christopher Miller, the Lead Analyst of Emerging Payments at Javelin Strategy & Research, looks at why this revolution has been slow to materialize.

“Just because 100 million people interacted with something, that’s not meaningful as a metric for understanding how quickly that technology will have an impact at the enterprise level,” Miller said.

Consumers Are Not Businesses

The message most people have heard about AI focuses on its record-setting adoption speed and the massive number of new users for ChatGPT. But that doesn’t always translate into enterprise use cases.

“It’s potentially interesting for understanding consumer interest in a technology or willingness to experiment with it,” said Miller. “The fact that someone was willing to download something or create an account is a holdover from consumer-focused app releases. It’s the kind of thing that you put in your pitch deck for VCs if you are launching a new mobile application. It is not the kind of thing that matters when you’re thinking about whether companies will do things in this or that way.”

Miller believes it will take roughly five years for AI to make a significant impact in the payments industry. So what’s holding it back?

Most obviously, concerns around accuracy and privacy remain major challenges for those developing generative AI solutions. Regulated firms often need to explain how decisions are made—making it risky to rely on tools when there’s uncertainty about how they arrive at their conclusions. Creating the legal, regulatory, and liability frameworks that allow businesses to confidently adopt a new technology like AI doesn’t happen overnight.

The Workflow Conundrum

There’s also concern around workflows. Miller describes generative AI as “automation on steroids,” helping people handle repetitive tasks that are embedded in their workflows.

“For example, if you needed to make a list of 10 companies and assemble some information about each of them, you could do 10 sets of Google searches,” Miller said. “Or you could use an AI prompt that makes a list of the 10 companies with their CEOs and their websites, and put it in a pretty chart and make it into a slide. You could theoretically do 500 of those a day instead of 10.”

The time savings from such projects would be minimal if the other processes in the workflow have not also enhanced their capabilities.

“Let’s say the review of your work is supposed to be done by a human being who is out of the office, so your work just sits there and waits for them to get back,” Miller said. “It doesn’t matter that you’ve become 50 times more productive, because the end product is not going to reach consumers any faster than it would have before. If everyone else along the line is not also brought along, no gain will be achieved and no change will actually happen.”

This issue persists even when the remaining bottlenecks are automated systems with inherent throughput limitations—whether they can process a limited number of documents at a time or were designed to accommodate a certain, expected flow of information.

“If we automate everything else, all of a sudden the flow that you need to handle is tripled, quadrupled, quintupled, centupled, whatever it would be,” Miller said. “These examples illustrate why it will take a long time to implement technologies in a way that will result in substantial gains that can actually be harvested.”

Vetting the Use Cases

It will take time for organizations to identify AI use cases that actually make a difference in customer sales or relationships. Many current generative AI success stories highlight impressive capabilities—but not all of them translate into sustainable or profitable outcomes.

“It is possible to generate unique marketing text for each of your clients,” Miller said. “That’s fascinating, and yet what is the outcome? Do we know that that level of customization will actually move the needle on whether the customer will buy anything or not? If the problem is that people don’t read the emails that you send to them, then having each of them contain unique text, while amazing from a technical standpoint, is immaterial from a business standpoint.”

A New Era?

Some are calling 2025 the beginning of a new era, driven by the rise of generative AI agents and their potential to reshape how consumers acquire financial products, make payments, and manage their finances. These shifts could eventually enable new payment rails, authentication methods, and monetization models—disrupting legacy products, companies, and practices. However, the underlying transformation necessary to support this evolution will take years to fully materialize.

“Any consumer-facing agent that you launch this year is likely to be garbage,” Miller said. “The only reason to do it is to learn from it, so it’s probably best done as small-scale pilots as you develop internal familiarity with the technology and its weaknesses. It gives you a way to track new models as they come out, but the product itself will not be a market differentiator for your service.”

While many enterprises are positioning AI at the center of their marketing strategies, these efforts may have limited impact on actual customer behavior—at least in the near term.

“They believe that adding AI constitutes part of their appeal or their pitch,” Miller said. “I don’t think that a lot of buyers are susceptible to that. People are not shopping for AI—they are shopping for increased speed or better productivity or improved customer experience. These are the things that enterprise shoppers are looking for.”